In order to analyze the social networks, several models have been proposed and studied in the literature. Toivonen et al. [1] made a comparison between some of these models using the last.fm and email network datasets. Since these datasets were not comprehensive enough, some defects of the presented models were not appeared in their comparison. In this paper, we will try to achieve a better comparison using the collected datasets from the facebook and Wiki-Vote networks.

© All Rights Reserved

22 tayangan

In order to analyze the social networks, several models have been proposed and studied in the literature. Toivonen et al. [1] made a comparison between some of these models using the last.fm and email network datasets. Since these datasets were not comprehensive enough, some defects of the presented models were not appeared in their comparison. In this paper, we will try to achieve a better comparison using the collected datasets from the facebook and Wiki-Vote networks.

© All Rights Reserved

- Research Outreach Issue 101
- Feature Engineering in Hybrid Recommender Systems
- On Realizing Rough Set Algorithms with Apache Spark
- Identification of Musically Induced Emotion: A Machine Learning Based Approach
- T-79_4202_solutions_4
- Proposed Platform IPHCS for Predictive Analysis in Healthcare System by Using Big Data Techniques
- Performance Evaluation of Data Placement Structures for Big Data Warehouses
- Approaches on Personal Data Privacy Preserving in Cloud: A Survey
- Big Data Analysis with Query Optimization Results in Secure Cloud
- tmp5437.tmp
- Fuzzy Neighborhood Grid-Based DBSCAN Using Representative Points
- Sleep Stage Classification: Scalability Evaluations of Distributed Approaches
- Page Rank Calculator In Java
- Credibility Analysis for Tweets Written in Turkish by a Hybrid Method
- Degree Distribution of Real Biological Networks
- Application of Stochastic Simulations of Biological Networks Under Different Impulsive Scenarios
- Watts 1998 Collective
- Freedom designator for sketchup.docx
- 9.0 Graphs Full Set
- D1,L4 Graphs and Networks.ppt

Anda di halaman 1dari 14

Hanif Emamgholizadeh Mahdieh Hasheminezha

FarzanehNasr Esfahani

Ali Dolati

Yazd University

h.emamgholizadeh@gmail.com

hasheminezhad@yazd.ac.ir

farzane.nasresfahani@gmail.com

adolati@yazd.ac.ir

ABSTRACT

In order to analyze the social networks, several

models have been proposed and studied in the literature. Toivonen et al. [1] made a comparison between some of these models using the last.fm and

email network datasets. Since these datasets were

not comprehensive enough, some defects of the presented models were not appeared in their comparison. In this paper, we will try to achieve a better comparison using the collected datasets from the

facebook and Wiki-Vote networks.

KEYWORDS

social network analysis, model, degree

distribution, average neighbors degree, geodesic

path length

INTRODUCTION

length and large clustering coefficient), powerlaw degree distribution, inverse relation between clustering coefficient and vertex degree,

and being assortative are the most important

properties for most of social networks. These

properties are known as "signature of social

networks". Most of analyzed social network

graphs have these properties. Evolution of presented models show that the researchers in this

field have tried to present some models to simulate the mentioned properties.

Social network analysis has some basic elements include: Average degree, clustering coefficient, geodesic path length and assortativity. Average degree in an indirected graph is

defined as:

k =

2 #Number o f links

.

#Number o f vertices

(1)

Wasserman [2] as average clustering coefficient of vertices, and for each vertex clustering

coefficient can be defined as:

C(G) =

3 #Number o f triangles

#Number o f

(2)

Geodesic path length between a couple of vertices is shortest distance between that pair of

vertices in graph, and the assortativity coefficient is the Pearson correlation coefficient of

degree between pairs of linked nodes. In a assortative graph vertices with large degree connect to other vertices with large degree. If average number of neighbors degree increase with

degree of vertices, graph is assortative.

The underpinnings of social network evolution

are to be found in modeling and studying evolving graphs. Informally, evolution refers to a

change that manifests itself across the time axis

[3]. For understanding dynamics of social networks we need to model them. Designing a realistic model, which can describe mechanism

of social network evolution is one of the challenging task in context of social network.

In 2009, Toivonen et al., [1] have compared the

proposed models for social networks with real

data. In their comparison, they used datasets

collected from last.fm website and email network. These two datasets have fairly small

clustering coefficient(between 0.2 and 0.3) and

small average degree (fewer than 10).

Choosing these data had two problems. First,

the social networks, from which these data

were collected, are not comprehensive, i.e.

these social networks have special users with

special behavior. The First problem causes

the second problem. Comprehensive social

22

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

large average degree. For example three comprehensive social network datasets were collected from facebook, twitter, and google plus

have clustering coefficients, more than 0.5, and

average degree more than 43 [4].

In this paper first, the web site used for collecting real datasets are introduced . then, the proposed models are explained briefly and finally

the modes are compared.

Both of datasets naturally have directed links.

Since in the modeling existance of relations

is important, not the nature of relations, we

consider these graphs as indirected networks.

And we will think about these directed links as

interaction between two users.

3.1 Categorizing

REAL DATA

is taken from [4]. Facebook is a comprehensive social network with large clustering coefficient, and Wiki-Vote has small clustering coefficient. Clustering coefficient of Wiki-Vote

was smaller than last.fm and email network, but

it has still small world property. This gap capable us to reveal the models defects. Defects

Which were not transpired in Toivonen's comparison.

In facebook graph nodes correspond to facebook users and edges are the friendship relations between these users. Wiki-Vote graph is

a subgraph of wikipedia network. In this graph

nodes show wikipedia users and edges is made

by a voting system. A user vote to other user if

he/she like that user's article, and by this way an

edge emerges between these two users. These

interactions make Wiki-Vote's social network.

We will describe these graphs in next subsection.

2.1

with more than one billion users. Each user

in facebook is considered as a vertex, and

friendship relation of two users is considered

as link connecting their coresponding vertices.

Date for Wiki-Vote has been collected from

wikipedia forum. Some users produce contents in forum and some other users vote to

his contents and some users with more votes

are chosen as forum administrator. Nodes in

the network represent wikipedia users and a

directed edge from node i to node j represents

introduced by Toivonen et al. [1] for grouping social network models. Toivonen et al.

divided social network models to two main

groups: network evolution models (NEMs),

and nodal attribute models (NAMs), and they

also divided NEMs to two subgroups: dynamical network evolution models (dynamical

NEMs), and growing network evolution models (growing NEMs). They showed each of

these groups appear different behaviors.

Network evolution models have three primitive

characteristics:

1. They change graph structure by repeating

a special process on graph with adding and

removing links and vertices.

2. In each time step they choose a subset

of links or vertices, and remove some of

these links or vertices by stochastic rules.

3. Each subgroups of NEMs has its own

termination criterion. Growing NEMs

stop when they reach to a determined

number of vertices, and dynamical NEMs

stop when their graphs reach to a stable

statistical status.

Growing network evolution models usually

start with an empty graph, and in each time

step, add a new vertex to the graph, or draw

some links between existing vertices. They iterate these process to reach a predetermined

number of vertices.

Dynamical network evolution models start

23

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

group

Dynamical NEMs

Growing NEMs

NAMs

Model

DEB

MVS

KOSKK

Vaz

TOSHK

BPDA

WPR

with a graph which has predetermined number of vertices, and in each time step remove

a number of vertices (and immediately add the

same number of vertices) or links, and also add

a number of links. Models iterate these to reach

a stable statistical status.

Nodal attribute models, add linke between each

pair of nodes, i.e. i and j, based on characteristics of vertices i and j.

Table 1 shows category of social network models. In subsequent subsection we introduce

these models.

3.2

Algorithm of DEB

by Davidsen et al [5] in 2002. They would

like to produce graphs with small world behavior. Algorithm 1 shows how this model makes

graphs. DEB has just one free parameter i.e.

probability of remove (p).

initialization: Make a graph with N

isolated vertices ;

step := 0 ;

while Graph is not in a stable status do

step := step + 1;

choose randomly a vertex named i ;

if i has fewer than 2 neighbors then

choose another vertex randomly

named j;

draw a link between i and j;

else

choose two neighbors of i named j

and k;

draw a link between j and k ;

end

choose randomly a vertex r with

probability p and remove r with all of

links;

end

In 2003, Vazquez presented a couple of models which produce graphs with local rules [6].

Only one of those models, which called Connecting nearest-neighbor model have social

network signature in produced graphs. In this

model there is only one free parameter u, which

is the probability of adding new vertex. Algorithm 2 shows how Vaz model works.

initialization: make graph with only one

vertex ;

step := 0;

NumberOfNodes := 1 ;

while NumberOfNodes N do

if RandomNumber u then

add a new vertex i ;

NumberOfNodes := NumberOfNodes + 1;

draw a link between i and another

randomly chosen vertex j;

create potential link between i and

neighbors of j these links are not

real links ;

else

change a potential link to real link

end

end

Boguna et al. proposed BPDA model in 2003

[7]. Producing a graph with high cluster coefficient, degree-degree correlation, and community structure 1 , by homophyly principle was

their goals.

Homophyly principle at first established by

McPherson et al. [10]. Essence of this principle summed up in this sentence: ''birds of a

feather flock together''. Homophyly principle

tells similar persons will be friend with more

probability.

In this model, vertices at first scattered in onedimensional segment [0, hmax ] and based on

their Euclidean distance link drawn between

them. Algorithm 3 showes that this model has

two free parameters and b. Parameter is

1 Communities

more inter community link than intra community links

for more read refer to [9]

24

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

average degree of graphs. We should choose

such a big number as hmax which probability of

vertices overlapping be very small, and increasing hmax , could not affect on resulting graph.

initialization: Scattered uniformly

vertices in one-dimensional segment [0, hmax ];

d is Euclidean distance brtween i and j ;

forall the pair of nodes do

draw link between i and j with

1

probability p = (1+(d/b))

end

3.5

Algorithm of MVS

et al. [8]. They tried to find an unstable state

in graphs which produced by this model. This

model has 3 free parameters: probability of

general search, probability of local search,

and probability of remove. Marsili et al.

showed in a special range of / , average degree and clustering coefficient of the model

have unusual behaviors. Algorithm 4 is the algorithm of MVS.

initialization: make a graph with N

vertices ;

step := 0 ;

while Graph is not in a stable status do

step := step + 1;

forall the vertices do

choose randomly another vertex j and

draw a link between current vertex

and j with probability ;

with probability draw a link

between current vertex and one of

the friends of friends which is

selected randomly ;

end

delete each of the links with

probability ;

end

3.6

realistic communities Toivonen et al. presented

this model.

In principle, this model has three free parameters. One of them is the number of selected

vertices for local search i.e. mr . This parameter

determined before starting the algorithm. Two

other parameters are ms , number of secondary

search and p, probability of choosing one of the

numbers 1 . . . mr (Algorithm 5).

initialization: make a graph with one

vertex;

NumberOfNodes :=1;

while NumberOfNodes N do

add node j to the graph;

NumberOfNodes := NumberOfNodes + 1;

choose 1 i mr with probability pi ;

choose i vertices of graph randomly;

n2nd U[1, k], k = 1, . . . , ms ;

choose n2nd number of neighbors of mr

chosen vertices;

draw links between current vertex and

these vertices;

end

Wong et al. in 2006 presented another model

based on homophyly principle [12], which is

known as WPR. There is some differences between BPDA and WPR. WPR distribute vertices in 2-dimensional space unlike BPDA, that

scattered vertices uniformly in 1-dimensional

space. WPR scattered vertices according to a

homogeneous Poisson point process on space

(see [13] [14]).

There is four free parameters in this model: H,

p, pb , p . Parameter H is the neighbourhood

radius, p is the average density of the network,

pb is the proximity bias which specifies the sensitivity to geographical space by the actors on

establishing social links, and p maintain the

expected average density of graph. WPR is

given in Algorithm 6.

Algorithm of TOSHK

konwn as TOSHK [11]. TOSHK uses local

rules for producing graphs. To make a graph

Kumpula et al. in 2007 proposed KOSKK

in [15]. This model was proposed to model

25

2-dimensional space according to a

homogeneous Poisson point process;

forall the pair of vertices do

if Euclidean distance is less than H then

draw link between current vertices

with probability p + pb ;

else

draw link between current vertices

with probability p p ;

end

end

graphs with weighted links. Some of the network characteristics, in addition to interactions, depends on weights of these interactions. KOSKK draw new links according to

the weights of links incident to a vertex. This

model has tree free parameters. Parameter p is

probability of local search, pg is probability of

general search and pr is probability of removing the node (Algorithm 7).

4

MODEL ANALYSING

produced by the models. But before, we should

fit parameters to produce graphs, which have

same value for some properties with real data.

After this fitting, we compare other properties

of graphs.

4.1

are: average degree of the graph, clustering coefficient, assortivity and average geodesic path.

In this paper we will use this priority.

As mentioned DEB and Vaz have only one

free parameter so we just can fit this parameter

to simulate first property i.e. average degree.

This does not mean that other property could

not be matched with real data, but means that

we have not any tools to control other properties.

Some of the other models have two or three

parameters, which can control clustering coefficient in addition to average degree. BPDA,

MVS and TOSHK are in this group of models.

isolate node;

set and W0 ;

while Graph is not in a stable status do

forall the vertices i do

if current vertex has at least one

neighbor then

choose one of i's neighbors j

according weigth of j which is

the summation of the weigth of

the incident links;

add to the weight of link {i, j};

choose one of the j's neighbors

k according weigth of k;

add to weight of link { j, k};

if there is a link between i and k

then

add to the weight of link

{i, k};

else

with probability p draw a

link between i and k, set the

weight {i, k} weight W0 ;

end

end

if current vertex has at most one

vertex, or with probability pg then

choose randomly another vertex j;

if there is a link between i and j

then

add to the weight of link

{i, j};

else

draw a link between i and j

and set the weight of link

{i, j} weight W0 ;

end

end

end

end

26

KOSKK and WPR have some other free parameters, that can be used to control some

other properties. KOSKK has which can

be fitted to simulate assortative and community structure properties. This parameter varies

between 0 and 1. When it approaches to 0,

assortativity coefficient increase, but community structure is destroyed in resulting graphs.

When approaches to 1, assortativity coefficient decreases and community structure is

made. Community structure is one of the most

important feature of social networks. When

approaches to 1 assortativity coefficient decreases but graph still remained assortative.

Toivonen et al. [1] set = 0.5 and with this

value for , they obtained very small measure

for assortative coefficient. Notwithstanding we

wont analyze community structure in this paper

but to produce graphs with community structure (our real data are so clustered) we increase

to 0.85. With this value for , our resulting

graphs are assortative. For WPR we fit parameter to produce graphs with average geodesic

path length similar to real data. Table 2 showes

fitted model parameters.

Results are averaged over 20 realizations of

each network model and come on Tables 3 and

4. In the following section we take a closer

look at this results. We produce graphs with

k = 43.69 0.5 and k = 29.14 0.5 to

model facebook and Wiki-Vote graphs respectively.

4.2 DEB analyzing

As we mentioned in the preceding section, this

model has only one free parameter i.e. probability of remove, and for this reason, we do not

have any control on clustering coefficient. We

can see this result in Tables 3 and 4. Clustering

coefficients of this model are not matched with

real data. This problem also exist in Toivonen

et al [1] results.

Initial purpose of this model was producing

small-world graphs. So resulting graph should

have small average geodesic path length. This

is obvious in Tables 3 and 4. This model's average geodesic path length is even smaller than

the real data.

4.3

Vaz analyzing

(and again like DEB) we only can control average degree of produced graphs. Tables 3

and 4 show, properties of this model. Average geodesic path length of this model is

small. For facebook graph this model graphs

geodesic path length is smaller, but for WikiVote this property is bigger than real data. With

up together these results and Toivonenet et al.

[1] results, we can conclude that for graphs

with small average degree, diameter of graph

is large but for dense graph, the diameter becomes smaller.

4.4

BPDA analyzing

some defects in modeling of social networks

which were not appear in Toivonen et al.

comparison. In the remaining models, we have

more than one free parameters, so we should

control some other properties, in addition to

average degree.

Cliques (fully connected subgraphs) are one of

the most important features in social networks.

Among all of the cilque types, triangle is the

most important. Being friend with friends of

friends is common relations in social networks

and providing models, for simulating this

feature of real data is one of the most important

task of researchers.

Before revolution of online social networks

like facebook, twitter and google plus online

social networks have small clustering coefficient. But These days dense social networks

with large clustering coefficient have emerged

(see collected data in [4]). For this purpose

we want to show how these models are not

efficient enough, and why we need new models

for modeling new social networks.

BPDA was first nodal attribute model which

proposed. This model in Toivonenet et al.

[1] comparison produced graph which its

clustering coefficient were almost matched

with real data's clustering coefficients. For

graphs which model's parameters fitted to

facebook graph this model, unlike the DEB

27

Name

DEB

Vaz

BPDA

MVS

TOSHK

WPR

KOSKK

Real data

facebook

Wiki-Vote

facebook

Wiki-Vote

facebook

Wiki-Vote

facebook

Wiki-Vote

facebook

Wiki-Vote

facebook

Wiki-Vote

facebook

Wiki-Vote

Parameters

p = 0.011

p = 0.016

u = 0.966

u = 0.945

= 44, = 3, d = [1, 11000]

= 320, = 1, d = [1, 70000]

= 0.0005, = 0.0934, = 0.004

= 0.0015, = 0.06640, = 0.0043

p1 = 0.999, p2 = 0.001, k = 189

p1 = 0.01, p2 = 0.01, p3 = 0.75, p4 = 0.23, k = 7

p = 0.07018, pb = 0.57, p = 0.06948, H = 2

p = 0.0685, pb = 0.08219, p = 0.068, H = 2

p = 0.11, pg = 106 , pr = 0.001

p = 0.09, pg = 0.06, pr = 0.002

Table 3. Basic statistics of the facebook sets and the models fitted to each. Results are averaged over 20 realizations of

each network model

Name

Facebook

DEB

Vaz

BPDA

MVS

TOSHK

WPR

KOSKK

Number of nodes

4039

4039

4039

4039

4039

4039

4039

4039

Average degree

43.69

43.9

43.52

43.19

43.98

43.97

43.56

43.83

Clustering coefficient

0.6

0.38

0.34

0.61

0.035

0.35

0.59

0.22

coefficient matches with facebook graph's

clustering coefficient (Table 3). It seems that

this model produces graphs with a wide range

of clustering coefficients, but this is not true,

as you can see in Table 4, clustering coefficient

of graphs fitted to Wiki-Vote is larger than

Wiki-Vote's graph clustering coefficients. We

tried to produce graphs, whose clustering

coefficient match with Wiki-Vote but, we

failed .

In this model we have two free parameters

and b. Parameter b is used for matching

average degree, thus, we have for controlling

clustering coefficient of produced graph. As

you can see in Table 2 for produced graphs,

similar to Wiki-Vote graph we choose = 1

i.e. smallest value. Even with this parameter

we could not produce graphs with clustering

Redious

5

4

4.7

39.5

3

4

4

1.3

Diameter

8

6.3

8.8

78.6

4

7.3

5.5

7.4

3.93

2.78

3.66

26.76

2.607

2.97

3.06

2.73

can claim, there is a lower bound for BPDA's

clustering coefficient spectra.

Having lower bound for clustering coefficient

spectra is not the only problem of this model.

As you can see in Tables 3 and 4, radius ,

diameter and average geodesic path length of

this model is considerably more than real data.

This has been shown in Toivonen comparison

too. In facebook graph three vertices with most

degrees have 1045, 792 and 755 neighbors,

but in BPDA's graphs (averaged over 20

realizations), vertices with most degrees have

only 66, 64.8 and 63.5 neighbors. We know

that hubs make short cut between vertices

and decrease their smallest path length [16]

[17]. Random graphs has vertices with small

degree and as Wang and Chen have shown

in [16] random graphs have somewhat larger

28

Table 4. Basic statistics of the Wiki-Vote sets and the models fitted to each. Results are averaged over 20 realizations of

each network model

Name

Wiki-Vote

DEB

Vaz

BPDA

MVS

TOSHK

WPR

KOSKK

Number of nodes

7115

7115

7115

7115

7115

7115

7115

7115

Average degree

29.14

29.01

28.78

29.05

29.24

28.79

28.9

29.43

Clustering coefficient

0.14

0.36

0.32

0.154

0.04

0.14

0.14

0.14

networks, and scale-free means emergence of

hubs. In following, we show this model has

other similarities to random graphs.

4.5

MVS analyzing

et al [1] claim that for small average degree of

graph this model could not produce large number of cliques, this is true according to the theoretical and numerical analysis in [8]. But there

is another problem, when average degree becomes larger, after a threshold , the larger average degree means the smaller clustering coefficients. Asymptotically clustering coefficient

for large average degree convergence to a very

small number (maybe less than 0.01).

Our graphs have large graph average degree,

so we expect clustering coefficient to be very

small. This is true according to numerical results in Tables 3 and 4. Toivonen et al [1]

results for clustering coefficient was match to

real data results because, their data had small

average degree (under the threshold ), and as

we showed this cluster even could not have

small world behavior (because its small clustering coefficients). MVS have small average geodesic path length even smaller than real

data.

4.6

TOSHK analyzing

we should increase p1 in one side, and decrease p2 in the other side. For example to

Redious

1

3.8

5.4

14

4

3

4

3.4

Diameter

7

7.4

10.4

28

4

5.3

5.6

6.3

3.24

3.15

4.02

10.04

2.94

2.95

3.31

3.07

we use from p1 we make two new triangles,

but with p2 we can not make any triangle. So

we enlarged p1 as far as we could, nevertheless

clustering coefficient only reached to 0.35 for

graphs similar to facebook real data.

Toivonen et al. [1] mentioned, TOSHK is unable to produce graphs with small clustering

coefficients. This is true if we confine ourselves to implemented version of TOSHK in

[11]. Author in implemented version [11] confined themselves to at most two general search

when new vertex is added i.e. mr = 2. But as

it can be seen in Algorithm 5, TOSHK algorithm is so flexible. When we increase number

of general searches we could reach to smaller

clustering coefficient, and truly, this is what we

did. We set mr = 4 and reached to our purpose.

For modeling Wiki-Vote by TOSHK we produce graph, which clustering coefficient were

0.14, equal with Wiki-Vote clustering coefficient.

As we can see in Tables 3 and 4, Average

geodesic path length of this model is smaller

than real data.

4.7 WPR analyzing

So far none of the discussed models could produce wide range of clustering coefficient spectra. But this model can produce graph with

such feature. As it can be seen in Tables 3 and 4

this model produce graphs with large clustering

coefficient (equal with real data). This model

geodesic path, also is very small number and

near to real data.

29

4.8

KOSKK analyzing

model. This model like MVS after a threshold of average degree clustering start to diminish. For instance for small average degree (under 10) clustering coefficient arise up to 0.67

but for large average degree it becomes smaller.

For this reason, Toivonen et al. had acceptable results. For our comparison when we try

to model facebook graph our clustering coefficient reach to 0.22. We take general search

probability a very small number (Table 2) but

even this small number could not remedy our

probelm.

For modeling Wiki-Vote, Which had small

clustering coefficient, there is not any problem,

this is obvious in Table 4. This model has small

geodesic average path length.

5

PLOTS OF MODELS

As Mislove [18] and Ahn [19] mentioned, Social networks have special behavior for degree

distribution, Clustering coefficient as a function of the vertex degree and Average nearestneighbor degree as a function of the vertex

degree. Social networks have power-law degree distribution. There is inverse relation between clustering coefficient and degree of vertices and for social networks average nearestneighbor degree of a vertex increase by its degree. In this section we compare this behavior

of models with real data.

5.1

Degree distribution

with preferential attachment 2 power-law destribution emerges. This reflects in our models, All the models which have preferential attachment (except MVS) have tail between poisson distribtion and power-law distribution (Figures 1(b), 1(c), 1(f) and 1(h)) as mentioned by

Tiovonen et al. [1]. NAMs models tails decay like poisson distribution (Figure 1(d) and

2 preferential

richer, pereferantial attachment says that the likehood of

a node being attachment to new link is proportional to

the node degree

absence of vertices with high degree prove this

issue. Although MVS have preferential attachment as local search but his distribution is similar to poisson distribtion (Figure 1(e)) it seems

link removing play an important role in degree

distribution of MVS.

5.2 Clustering spectrum

Clustering spectrum C(k) of the Wiki-Vote

data and models fitted to it is shown in Figure

2. As mentioned, social networks clustering

coefficient have inverse relation with degree

of verticies. All of NEMs have this behavior

(Figures 2(b), 2(c), 2(e), 2(f) and 2(h)).

Toivonen et al. showed that for NAMs there

is a constant relation between clustering

coefficient and degree of verticies, this is true

for BPDA (Figure 2(d)) but our result for

WPR (2(g)) showes direct relation between

clustering coefficient and degree of vertices.

Maybe this difference comes from homogeneous Poisson point process. In our fitted

models (for facebook and Wiki-Vote), poisson

distribution mean and neighbourhood radius

were 7.5 and 2 respectively (facebook fitted

version of WPR has direct relation too which

has not shown in Figure), but Tiovonen et al.

[1] neighbourhood radius was less than 0.05.

This maybe cause to this difference. Another

probability is this, with increasing average

degree of graph, direct relation arise. Because

there is no theoretical proof of these claims in

hand, so more numerical simulation needs to

prove this claims.

Average nearest-neighbor degree as a function

of the vertex degree for facebook and models

fitted to it, is shown in Figure 3. As you can

see in Figures 3(b), 3(c), 3(d), 3(f), 3(g) and

3(h) except MVS which is almost disassortative, all of the models are assortative and have

same behavior with real data. Toivonen et al.

mentioned that this behavior of MVS return to

his nature i.e adding and deletion of nodes.

It is good to recall, KOSKK could have more

30

Wiki-Vote

DEB

Vaz

BPDA

MVS

TOSHK

WPR

KOSKK

Figure 1. Degree distributions P(k) of the Wiki-Vote data and models fitted to it. Results are averaged over 20 realizations

of each network model fitted to Wiki-Vote graph

31

c(k) in Wiki-Vote

(b) Clustering

c(k) in DEB

spectrum

(c) Clustering

c(k) in Vaz

spectrum

(d) Clustering

c(k) in BPDA

spectrum

(e) Clustering

c(k) in MVS

spectrum

(f) Clustering

c(k) in TOSHK

spectrum

(g) Clustering

c(k) in WPR

spectrum

(h) Clustering

c(k) in KOSKK

spectrum

Figure 2. Clustering spectrum c(k) for the Wiki-Vote and data models fitted to it. Results are averaged over 20 realizations

of each network model fitted to Wiki-Vote graph.

32

(a) Average neighbors de- (b) Average neighbors degree Knn in Wiki-Vote

gree Knn in DEB

Figure 3. Average neighbors degree Knn for the facebook data and models fitted to it. Results are averaged over 20

realizations of each network model fitted to face book graph.

33

to have less steep.

6

CONCLUSION

could not produce all behavior and features,

i.e. signature, of social networks. Some

could not produce graph with large clustering

coefficient and others do not have power-law

degree distribution. Some have direct relation

between clustering coefficient and degree of

verticies and others are not assortative.

We had a condense and comprehensive glimpse

on proposed models for social network, and

reveal latent aspects of these model, which

neglected in previous comparison. We showed

that more work is needed in this area. Up

together this result, show we need some new

models, for modeling new social networks,

networks which become denser and more

clustered.

ACKNOWLEDGMENT

We would link to thanks to Mr. Meysam Rajaati, Mr. Hadi Abdivash, Mr. Mojtaba Jafari

Qamsari and Mr. Ali Shkiba for their helps.

We acknowledgment the Laboratory of Quantum Computing of yazd university.

Modeling acquaintance networks." Physical Review Letters 88, no. 12 (2002): 128701.

[6] A. Vazquez. "Growing network with local rules:

Preferential attachment, clustering hierarchy, and

degree correlations." Physical Review E 67, no. 5

(2003): 056104.

[7] Boguna, Marian, Romualdo Pastor-Satorras, Albert Diaz-Guilera, and Alex Arenas. "Emergence

of clustering, correlations, and communities in

a social network model." arXiv preprint condmat/0309263 (2003).

[8] M. Marsili, F. Vega-Redondo, and F. Slanina. "The

rise and fall of a networked society: A formal

model." Proceedings of the National Academy of

Sciences of the United States of America 101, no.

6 (2004): 1439-1442.

[9] S. Fortunato. "Community detection in graphs."

Physics Reports 486, no. 3 (2010): 75-174.

[10] M. McPherson, L. Smith-Lovin, and J. M. Cook.

"Birds of a feather: Homophily in social networks." Annual review of sociology (2001): 415444.

[11] R. Toivonen, J. Onnela, J. Saramaki, J. Hyvonen,

and K. Kaski. "A model for social networks." Physica A: Statistical Mechanics and its Applications

371, no. 2 (2006): 851-860.

[12] L. H. Wong, P. Pattison, and G. Robins. "A spatial

model for social networks." Physica A: Statistical

Mechanics and its Applications 360, no. 1 (2006):

99-120.

REFERENCES

[1] R. Toivonen, L. Kovanen, M. Kivela, j. Onnela, J.

Saramaki, and K. Kaski. "A comparative study of

social network models: Network evolution models

and nodal attribute models." Social Networks 31,

no. 4 (2009): 240-254.

[2] S. Wasserman. Social network analysis: Methods and applications. Vol. 8. Cambridge university

press, 1994.

[3] M. Spiliopoulou. "Evolution in social networks: A

survey." Social Network Data Analytics. Springer

US, (2011). 149-175.

[4] J. Leskovec. "Stanford large network dataset collection. snap." (2010).

[5] J. Davidsen, H. Ebel, and S. Bornholdt. "Emergence of a small world from local interactions:

[13] D. R. Cox, and V. Isham. "Point Processes (Monographs on Applied Probability and Statistics)."

Chapman and Hall, London(1980).

[14] J. F. Kingman. "Poisson processes, volume 3 of

Oxford Studies in Probability." The Clarendon

Press Oxford University Press, New York (1993).

[15] J. M. Kumpula, J. Onnela, J. Saramaki, K. Kaski,

and J. Kertesz. "Emergence of communities in

weighted networks." Physical review letters 99, no.

22 (2007): 228701.

[16] X. F. Wang, and G. Chen. "Complex networks:

small-world, scale-free and beyond." Circuits and

Systems Magazine, IEEE 3, no. 1 (2003): 6-20.

[17] A-L. Barabasi, and R. Albert. "Emergence of scaling in random networks." science 286, no. 5439

(1999): 509-512.

34

and B. Bhattacharjee. "Measurement and analysis

of online social networks." In Proceedings of the

7th ACM SIGCOMM conference on Internet measurement, pp. 29-42. ACM, 2007.

of huge online social networking services." In

Proceedings of the 16th international conference

on World Wide Web, pp. 835-844. ACM, 2007.

35

- Research Outreach Issue 101Diunggah olehResearch Outreach
- Feature Engineering in Hybrid Recommender SystemsDiunggah olehliezelleann
- On Realizing Rough Set Algorithms with Apache SparkDiunggah olehliezelleann
- Identification of Musically Induced Emotion: A Machine Learning Based ApproachDiunggah olehliezelleann
- T-79_4202_solutions_4Diunggah olehbobopower
- Proposed Platform IPHCS for Predictive Analysis in Healthcare System by Using Big Data TechniquesDiunggah olehliezelleann
- Performance Evaluation of Data Placement Structures for Big Data WarehousesDiunggah olehliezelleann
- Approaches on Personal Data Privacy Preserving in Cloud: A SurveyDiunggah olehliezelleann
- Big Data Analysis with Query Optimization Results in Secure CloudDiunggah olehliezelleann
- tmp5437.tmpDiunggah olehFrontiers
- Fuzzy Neighborhood Grid-Based DBSCAN Using Representative PointsDiunggah olehliezelleann
- Sleep Stage Classification: Scalability Evaluations of Distributed ApproachesDiunggah olehliezelleann
- Page Rank Calculator In JavaDiunggah olehParvezAhmed Kazi
- Credibility Analysis for Tweets Written in Turkish by a Hybrid MethodDiunggah olehliezelleann
- Degree Distribution of Real Biological NetworksDiunggah olehliezelleann
- Application of Stochastic Simulations of Biological Networks Under Different Impulsive ScenariosDiunggah olehliezelleann
- Watts 1998 CollectiveDiunggah olehpastafarianboy
- Freedom designator for sketchup.docxDiunggah olehpedrokalungsod
- 9.0 Graphs Full SetDiunggah olehFaiz Ramli
- D1,L4 Graphs and Networks.pptDiunggah olehmokhtarppg
- CounterexamplesDiunggah olehdragance106
- Ws Dm 12 TutorialDiunggah olehAmel Ksibi
- NatSci Manuscript.pdfDiunggah olehPiraveenan Mahendrarajah
- 20160826 Incidence Matrix SupplDiunggah olehJoyprakash Lairenlakpam
- Itute 2008 Exam 1Diunggah olehmishiemichelle
- Pathways Meaning Text Network AnalysisDiunggah olehjsbarandiaran
- 0352-51390503441GDiunggah olehmvargasl
- 46840_PDKU- 2008 Tomescu DM Discrepancies Metric and Partition DimensionsDiunggah olehBaiq Supriani
- DbsDiunggah olehduyhoa83
- adhocDiunggah olehrajouta19

- Strengths and Vulnerabilities of Cloud Computing in MauritiusDiunggah olehliezelleann
- Fuzzy Neighborhood Grid-Based DBSCAN Using Representative PointsDiunggah olehliezelleann
- Degree Distribution of Real Biological NetworksDiunggah olehliezelleann
- Performance Evaluation of Data Placement Structures for Big Data WarehousesDiunggah olehliezelleann
- Prototyping an Intuitive Circuit Simulator Manipulated by Hand Gestures for Elementary Physics EducationDiunggah olehliezelleann
- Analysis and Performance Evaluation of Convolutional Codes over Binary Symmetric Channel Using MATLABDiunggah olehliezelleann
- Application of Stochastic Simulations of Biological Networks Under Different Impulsive ScenariosDiunggah olehliezelleann
- Big Data Analysis with Query Optimization Results in Secure CloudDiunggah olehliezelleann
- A Study on the Stratospheric Environmental Factors in the Philippines that Affects the Quality of Data Gathered in Launching Near Space BalloonDiunggah olehliezelleann
- Proposed Platform IPHCS for Predictive Analysis in Healthcare System by Using Big Data TechniquesDiunggah olehliezelleann
- Sleep Stage Classification: Scalability Evaluations of Distributed ApproachesDiunggah olehliezelleann
- Dispersed Remote Vehicle Diagnostic SystemDiunggah olehliezelleann
- Toward Forensics Aware Web Design: Case Study: Low Fare Local Hajj Booking Web ServicesDiunggah olehliezelleann
- Approaches on Personal Data Privacy Preserving in Cloud: A SurveyDiunggah olehliezelleann
- Behavioural Models for Business Usage: The Case of Online Social Network UsersDiunggah olehliezelleann
- Development of a City Planning Simulation System in the AR SpaceDiunggah olehliezelleann
- Credibility Analysis for Tweets Written in Turkish by a Hybrid MethodDiunggah olehliezelleann
- Development of the Intuitive and Seamless Synthetic System of the Interior and Exterior in a Building by Using ARDiunggah olehliezelleann
- Block Based Technique for Detecting Copy-Move Digital Image Forgeries: Wavelet Transform and Zernike MomentsDiunggah olehliezelleann
- Microcontroller - Based Embedded System for Drowsiness Detection and Vibration AlertnessDiunggah olehliezelleann
- Time Synchronization Algorithm for Asymmetric Optical Fiber Communication LinkDiunggah olehliezelleann
- Forensic Network Traffic AnalysisDiunggah olehliezelleann
- Modeling and Simulation for Orientation Control of an Under-actuated Drill MachineDiunggah olehliezelleann
- Using Fuzzy Logic to Predict Winners in Horseraces at the Champ de MarsDiunggah olehliezelleann

- Exp 4Diunggah olehVickyBoo
- 19 JavaDiunggah olehNeha Chack
- Air Suspension OnDiunggah olehLokendra
- IND23-4V TrojanRE Data SheetsDiunggah olehJohn Melanathy II
- 3com Configuration GuideDiunggah olehMohammad Faizal Md Nor
- Advanced Java Lecture-7Diunggah olehcupidcallin
- 00510243Diunggah olehJorge Enrique
- AnyCasting.pdfDiunggah olehSimon Risteski
- A Channel Model for Power Line Communication Using 4PSK Technology for DiagnosDiunggah olehPriyanka Bhartiya Agarwal
- Ieee Format Seminar FormatDiunggah olehAmit Pandey
- ESM186L Lab ManualDiunggah olehangela131313
- Lab 7Diunggah olehzaq
- Management Scientist User's GuideDiunggah olehdonafutow2073
- Hilbert in My View - F.rotheDiunggah olehbeni
- ILF-SPC-SRT-EL-806-0 Electrical LV Switchgear - Specification.pdfDiunggah olehMaheshNayeni
- IIT Video Lectures on Active Filter Design by DrDiunggah olehsatishbojjawar1484
- Fluttering Thesis.pdfDiunggah olehInge Devriendt
- 130001-1 Basic Electronics gtu 3rd sem paperDiunggah olehShailesh Sankdasariya
- Handbook on Maintenance of Switch Expansion Joint(3)Diunggah olehfiemsabyasachi
- iec61649 (EXTRACTO)Diunggah olehGabriel Aravena Aguilar
- Datastage interview questionDiunggah olehchandrasekharch06
- Swami the Leg to Body Ratio as a Human Aesthetic CriterionDiunggah olehalexander_santos_6
- Impact TestDiunggah olehwaysh
- Linearity and Analog Performance Analysis of Double Gate Tunnel FET: Effect of Temperature and Gate StackDiunggah olehAnonymous e4UpOQEP
- Design of Flexible PavementDiunggah olehyashirup
- Indirect Direct Heaters ComparisonDiunggah olehankur2061
- Solutions Icda HwDiunggah olehFankaiZhou
- Triggers in BESADiunggah olehtobylj5227
- pgdump-pgrestoreDiunggah olehlaverdadciega
- Gate StructuresDiunggah olehgrkvani10